Web Browsing and Automation

Web Browsing and Automation

Use browser capabilities to read web pages, PDFs, search, and take screenshots.

Available Functions

# Read a webpage in text format
read_url(url: str) -> str

# Search for information
search(query: str, engine: Literal["google", "duckduckgo", "perplexity"]) -> str

# Take a screenshot of a webpage
screenshot_url(url: str, path: Optional[Path]) -> Path

# Read browser console logs
read_logs() -> str

Best Practices

1. Start with search for unknown URLs: Use search() to find relevant pages
2. Read before extracting: Use read_url() to see page content
3. Check logs for errors: Use read_logs() after reading complex pages
4. Screenshot for visual verification: Useful for checking layouts or UI

Common Patterns

Searching and Reading

# Search for information
search("gptme documentation getting started")

# Read a specific page
read_url("https://gptme.org/docs/getting-started.html")

Using Perplexity for Research

# Get AI-powered answers with sources
search("latest developments in LLM agents", "perplexity")

Note: Perplexity search requires either:

• PERPLEXITY_API_KEY - Direct access to Perplexity API
• OPENROUTER_API_KEY - Uses Perplexity via OpenRouter (model: perplexity/sonar-pro)

If both keys are available, PERPLEXITY_API_KEY takes precedence.

Taking Screenshots

# Screenshot and view
screenshot_url("https://gptme.org")

Reading PDFs

# Read arXiv paper (default: first 10 pages)
read_url("https://arxiv.org/pdf/2410.12361v2")

# Read more pages
read_url("https://arxiv.org/pdf/2410.12361v2", max_pages=20)

# Read all pages
read_url("https://example.com/document.pdf", max_pages=0)

Note: PDF support uses pypdf library for text extraction. PDFs are automatically detected by:

• URL ending in .pdf
• Content-Type header containing application/pdf

Vision Alternative: For scanned/complex PDFs with garbled text, convert to images and use vision.

Debugging Web Pages

# Read page
read_url("https://example.com")

# Check for JavaScript errors
read_logs()

When to Use Browser vs Direct Tools

• Use browser for: Web research, documentation, public information
• Use shell for: Local files, git operations, system commands
• Use ipython for: Data processing, calculations, API calls